Download as pdf
Download as pdf
You are on page 1of 11
"JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION VOL. 34,NO.1 "AMERICAN WATER RESOURCES ASSOCIATION FEBRUARY 1998 UNIT HYDROGRAPHS DERIVED FROM THE NASH MODEL! Michel C. Boufadel® ABSTRACT; Unit hydrograph ordinates are often estimated by deconvoluting excess rainfall pulses and corresponding direct rruneff. Tho resulting ordinates are given at discrete times spaced ‘evenly at intervals equal tothe duration of the rainfall pulse. Ifthe ‘new duration is not a multiple of the parent duration, hydrograph interpolation is required. Linear interpolation, piece-wise nonlinear interpolation and graphical smoothing have been used. These inter- polation schemes are expedient but they lack theoretical basis and can lead to undesirable results Tnterpolation can be avoided if the instantaneous unit hydro- ‘graph (IUH) forthe watershed is known. Here two iarves connected With the classic Nash TUH are examined: (1) how should the Nash parameters be estimated? and (2) under what conditions is the Ferulting hydrograph able to reasonably represent watershed response? In the firt case, nonlinear constrained optimization pro- ‘ides better estimates ofthe IUH parameters than does the method ‘of momenta. In the second case, the Nash TUH gives good results on ‘watersheds with mild shape unit hydrographs, but performs poorly fon watersheds having sharply peaked unit hydrographs. Overall, in ‘comparison to empirical interpolation alternatives, the Nash [UH offers a theoretically sound and practical approach to estimate unit hydrographs for a wide variety of watersheds. (REY TERMS: need 6.6 key words here for indexing and informa tion retrieval.) INTRODUCTION ‘The unit hydrograph concept, introduced by Sher- man (1932), remains one of the most practical tools for estimating rainfall runoff. The essence of unit hydrograph theory is embodied in the matrix equa- tion: [P-L (0) =(@-B) where P is total rainfall, L is rainfall loss, U is the unit hydrograph, Q is the total runoff hydrograph and 67/0522 1 B is base flow. In Equation (1), the quantity P-L rep- resents the excess rainfall hyetograph (ERH) while Q-Bis the direct runoff hydrograph (DRH). When the unit hydrograph is known, Equation (1) provides estimates of the direct runoff expected from any given pattern of excess rainfall. Or, if the direct runoff and the excess rainfall are known, Equation (1) can be solved to estimate the ordinates of the corre- sponding unit hydrograph. In the latter ease, the total number of direct runoff hydrograph ordinates often exceeds the total number of unit hydrograph ordi- nates. Because there are more equations than unknowns, this leads to an over-determined system of equations in U. The goal then is to find an optimum set of unit hydrograph ordinates that satisfies Equa- tion (1). Optimal unit hydrograph ordinates are usually estimated by mathematical programming. The method is typically applied to several rainfall-ranoft events (Hjelmfelt and Kramer, 1990). Horton’s and Philip's infiltration equations have been used to rep- resent losses (Bagleson et al., 1966; Newton and Vin- yard, 1967; Singh, 1976; Mays and Coles, 1980). In other cases, the losses and the unit hydrograph have been optimized simultaneously, assuming that both are unknown (Mays and Taur, 1982; Unver and Mays, 1984). When developing the unit hydrograph (UH) for a given watershed, the duration of the excess rainfall pulse, At, is assigned to the resulting unit hydro- graph. As a point of clarification, the duration of the ‘unit hydrograph refers to the time interval between hydrograph ordinates and not the base time of the unit hydrograph. A I-hour unit hydrograph may have, for example, a base of 12 hours during which sPaper No, 97001 of the Journal of the American Water Resources Association, Discussions are open until October 1, 1998. 2Greduate Research Assistant, Department of Civil and Environmental Engineering, University of Cincinnati, PO, Box 7, Cincinnati, (Ohio 45221-0071 (e-m: mboufade@boss.cee.ue.edu). JOURNAL OF THE AMERICAN WaTeR RESOURCES ASSOCIATION JAWRA Boufadet time each discrete unit hydrograph ordinate is sepa- rated by an interval of exactly one hour. In many rainfall runoff problems, the duration of the rainfall pulses may differ from the duration of the unit hydrograph. In some cases, the rainfall can be discretized so that pulse durations match the unit hydrograph. In other cases, though, it may be neces- sary to change the duration of the unit hydrograph 50 that it agrees with the duration of the rainfall pulse. Extending the duration of a unit hydrograph is done readily with the S-method (Chow et al., 1988; Singh, 1992) or the forward-backward S-method (Tauxe, 1978). Shortening the duration of a unit hydrograph, however, can lead to problems. In what follows, this paper illustrates typical prob- lems encountered when short duration unit hydro- graphs are derived from parents of longer duration. It demonstrates how these problems can be avoided with application of the instantaneous unit hydro- graph (IUH). Then focusing on the classic Nash IUH, this paper compares several techniques to estimate the IUH parameters and investigate the effect of hydrograph duration on these parameter estimates. Finally, using runoff data from five watersheds, it compares the performance of short duration hydro- graphs derived from the Nash IUH against short duration hydrographs obtained by other means. CHANGING HYDROGRAPH DURATION Suppose a “parent” unit hydrograph hy(t) of dura- tion At; for a watershed is available and we wish to obtain the unit hydrograph ha(t) of duration Aty. The discrete S-hydrograph is defined only at multiples of ty. Hence, if Atg is not an integer multiple of At, interpolation between hydrograph ordinates is need- ed. To illustrate, consider a situation where Atz ‘/2Aty. In order to apply the S-hydrograph method, we need information at times midway between the ordinates of the parent unit hydrograph. This infor- mation, however, is not normally available. ‘An example based on this linear interpolation scheme with Ate = 24te is shown in Figure 1. The par- ent unit hydrograph has four non-zero ordinates while the derived unit hydrograph has eight non-zero ordinates. Note, however, that the derived unit hydro- graph has a “stepped” appearance. The UH ordinates are connected by straight lines to emphasize that the rising and falling limbs of the derived unit hydro- graph appear interrupted by periods of constant flow. Although the linear interpolation approach conserves the mass of the hydrograph, the resulting stepped response is difficult to justify from a physical stand- point, — - Possible Shape of Parent UH 2 Estimated Short-Duration UH — Interporlated Parent UH © _ Known Points Unit Hydrograph (1/time) Figure 1. Stepped Unit Hydrograph Resulting from Linear Interpolation. Other hydrograph interpolation approaches have been proposed. Diskin (1969) used quadratic piece- wise interpolation to estimate unit hydrographs from discrete runoff data. Sangal (1986) suggested graphi- eal smoothing to obtain a short duration unit. hydro- graph from one of longer duration. There is, however, no theoretical reason to accept these schemes nor to prefer one approach over another. The scheme pre- sented by Diskin, (1969) depends on the UH duration whereas the scheme presented by Sangal (1986) depends on both UH duration and the person per- forming the graphical smoothing. Furthermore, nei- ther of the schemes has a physical basis. These shortcomings can be eliminated if the dura- tion of the rainfall pulse for the parent hydrograph vanishes. This situation requires estimation of the “instantaneous unit hydrograph” (IUH) for a water- shed. In contrast to the methods mentioned above, the TUH offers an objective and rigorous framework for estimating unit hydrographs of any desired duration. 168 JOURNAL OF THE AMERICAN WaTER ReSOURCES ASSOCIATION ‘Unit Hydrographs Derived from the Nash Model NASH IUH. ‘Nash (1957) showed that the IUH of watershed represented conceptually by a series of n identical lin- ear reservoirs is: @ where h(t) is the discharge per unit area at time t, k is the coefficient of storage of each linear reservoir, and I(n) is the standard gamma function: Tin)= [> m*tedm @ where m is an integration (or dummy) variable. The corresponding S-hydrograph for the series of n identi- cal linear reservoirs is found by integrating (2) a= [ade @ By virtue of Equations (2) and (3), Equation (4) can be rewritten as: «) where G(*;*) is the ratio of the incomplete gamma fanetion to the gamma function and is tabulated in Abramawitz and Stegun (1972). ‘The continuous S-hydrograph given in Equation (5) is explicitly defined for all values of t > 0. Hence, unit hydrographs of any specified duration can be obtained from the Nash IUH without interpolating hydrograph ordinates. It follows that the ordinates of a unit hydrograph of duration At, are given by: ) ‘To use this expression it is necessary only to estimate the parameter pair (n,k). The effects of these two parameters on the Nash IUH are investigated by Boufadel (1992) and will be discussed briefly later in this manuscript. Several conventional parameter esti- mation methods are reviewed in the next section. ‘JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION PARAMETER ESTIMATION ‘The Nash IUH given in Equation (2) is a gamma probability density function with parameters n and k. Based on theoretical consideration regarding the lin- earity of Equation (1), Nash (1957, 1958) presented the method of moments in a simplified form to esti- mate n and k from observations of excess rainfall of duration t and corresponding direct runoff: mx(@) - Hy(P) = nk mM 2(Q)- He(P) - nkpy(P) = n(n +1)k2 @) where 1,(Q) is the rth moment of the DRH about the time origin divided by total DRH, and 11,(P) is the rth moment of the ERH about the time origin divided by total ERH. The method of moments tends to give more “weight” to the extremities of a distribution than to the center and consequently the resulting TUH is more in error near the peak than at the extremities (Nash, 1957). In view of this, Nash sug- gested the rigorous least squares method (Beck and Arnold, 1977, pg. 117) as an alternate fitting tech- nique. Gray (1961) proposed using maximum likelihood to estimate n and k for a dimensionless IUH. Although maximum likelihood estimators for the gamma distri- bution are not unbiased for small samples, corrections for the bias are available (Haan, 1977). Recently, Hak- tanir and Sezen (1990) equated the peaks of simulat- ed and observed IUHs to estimate n and k. Their procedure gives a very rough approximation because it uses only one observation (the peak) from the hydrograph. In what follows, I demonstrate an improved parameter estimation approach based on nonlinear constrained optimization. EXPERIMENTAL METHOD A three phase experimental procedure was estab- lished to identify the conditions under which the Nash IUH provides a satisfactory method for deriving unit hydrographs of different durations. I obtained “parent” unit hydrographs from five watersheds in the United States. These hydrographs, summarized in Table 1, were scaled by their drainage area to facilitate direct comparison of results among the five locations. 169 JAWRA Boufadel ‘TABLE 1. Five Parent Unit Hydrographs. Unit Watershed ty ‘ata t Hydrograph Location (he) cS) ay Source of Data Bradford Ohio 198 60 120 60 ‘Miami Conservancy District Goodwater Missouri 469 os 20 Ww USDA.Agr Rea, Ser Potomac Maryland 194 40 80 6 ‘Mays and Coles, 1980 Shoal Creek ‘Teas 103 08 10 5 ‘Mays and Taur, 1962 Wille Crock Maryland mt 30 60 0 Unver and Mays, 1984 Phase 1: IUH Parameter Estimation ‘Two methods for estimating the Nash IUH parame- ters n and k were compared. In the first approach, n and k were estimated with the method of moments (Equations 7 and 8) as given by Nash (1958). In the second approach, n and k were found using nonlinear constrained optimization with a normalized objective function given by the weighted sum of absolute errors: Eh-ni 4% mh i (9) where n is the total number of hydrograph ordinates, hy = h(At) is the observed UH ordinate at time t = iat, the predicted UH ordinate, his the average of 1,2,. . .n), and w; is the weight assigned to = (hy - h*)). The weight function is taken from It has a lower limit of 0.5 and a mean of 1.0. The weight function gives more importance to the region near the hydrograph peak. Although it introduces a slight bias into the optimization process, the weight function reduces the oscillatory behavior observed in the tail of many optimized unit hydrographs from noise in the data (Eagleson et al., 1966; Blank et al., 1971). Combining Equations (6) and (9) gives the 1UH parameter estimation objective function: [dott (i= pat k laciaey——2 4 nS fall an JAWRA This expression was optimized with a generalized reduced gradient search scheme available in the soft- ware package GRG2 (Lasdon et al., 1978). Optimi: tion was carried out subject to a mass conservation constraint that limited the area under the estimated TUH to a certain percentage € of the theoretical maxi- mum. The mass conservation constraint was needed because each scaled unit hydrograph encloses an area equal to 1.0. With the Nash IUH model, this condition occurs only when the ratio of the base time of UH and the storage term k tends to infinity, a situation which is not practical from a physical point of view. In this study, € = 95 percent was selected. The performance of the fitted hydrographs obtained from the method of moments and from opti- mization were then compared against the original At parent hydrograph using three criteria. Criterion 1, Scaled sum of absolute errors (SAE), F;, F; is merely Equation (9) with the weights set to unity (or omitted). While a variable weight function was useful for optimization purposes, it is not needed when comparing goodness of fit among different. unit hydrographs. Criterion 2. Scaled difference in hydrograph peaks, Fa. F is the ratio of the difference between the observed and simulated (or fitted) peaks of the UH to the peak of the observed UH. Criterion 3. The third measure is a plot of the derived unit hydrograph. This is rather subjective, but it has the advantage of revealing the overall good: ness of fit. Although information about the unit hydrograph ordinates is unknown except at time steps of At, the UH ordinates are connected by straight lines for plotting purposes. Phase 2: Unit Hydrograph Duration The parameter pair (n,k) should be independent of hydrograph duration. This property was examined by 170 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION ‘Unit Hydrographs Derived from the Nash Model checking the sensitivity of parameter estimates to hydrograph duration. This approach took an inverse route, Using the S-method, unit hydrographs of a long duration At were generated from the five parent hydrographs of a short duration At;. The interval At ‘was taken to be an integer multiple (usually 2) of At. ‘Then, using only the UH ordinates given at the long time intervals At, the parent hydrographs of short duration At; were re-estimated using two approaches. ‘The first approach applied the Nash IUH with parameter pair (n,k) found by nonlinear constrained optimization. The second approach used linear inter- polation between UH ordinates and then applied the ‘S-method. Differences between the parent At; unit hydro- graphs and the simulated At, unit hydrographs result from two kinds of errors. First, there will be errors due to the conceptual model itself. These errors me sure the suitability of the Nash model to represent the IUH from the specified watershed. Second, there will be errors due to the change of duration of the unit hydrograph. These effects reflect the qua: behavior of the rainfall-runoff process (whereas Equa- tion 1 assumes a linear relation between rainfall and runoff) and natural noise in the hydrograph data ‘The resulting unit hydrographs of duration at, were ‘compared against the original At, parent hydrographs using the three criteria mentioned in phase 1. ‘n and k obtained from optimization during Phases 1 and 2 were compared (not shown) because they were expected to be independent of duration. For the most part, their values did not change much between phas- es 1 and 2. Likewise, the product nk, which measures the centroid of the unit hydrograph, was fairly consis- tent between both phases. Phase 3: Runoff Simulation Each of the five watersheds had four unit hydro- graphs: the parent and three others derived from it. ‘The three derived hydrographs included two generat- ed with the Nash IUH (from nonlinear enstrained optimization and from the method of moments) and one obtained by combining linear interpolation with the S-hydrograph method. In this third phase, a rain- fall-runoff analysis was performed to determine which of the three derived unit hydrographs best mimicked the total response produced by the parent. A hypo- thetical storm with 12 pulses was applied to each hydrograph at each watershed using the HEC-1 rai fall runoff model. The three criteria mentioned in Phase 1 were used to compare results obtained from the derived hydrographs against those obtained from the parent hydrograph. RESULTS OF COMPUTER EXPERIMENTS Phase I: IUH Parameter Estimation ‘As shown in Table 2, hydrographs with parameters estimated by nonlinear constrained optimization (GRG2) tended to fit the parent hydrograph better than hydrographs derived from the method of moments (MoM). This is not too surprising since shortcomings with the method of moments have already been noted (Nash, 1957). Practically speak- ing, GRG2 equaled or outperformed MoM in minimiz- ing both F; (sum of absolute errors) and F» (difference between peaks) at all watersheds except at Wills Creek. In the case of Wills Creek, MoM got slightly ‘TABLE 2. Results of Phase 1 and Phase 2 Hydrograph Estimation —__ renee — rr eFrl—=$=wananpoaoao Phase 1: TUH Parameter Phase 2: Effect of Unit, Estimation Methods Hydrograph Duration Unit Fy @ercent) F, (percent) F; (percent) Fy (percent) Hydrograph GRG2 MoM GaGa ‘MoM GRc2 MoM = GRG2— MoM Bradford 94 186 0.008 “ana 9.53 60 0.35, 327 Goodwater 136 238 0.43 118 163 190 4.90 129 Potomac 137 wa 3.08 892 1a 184 2.36 9.48 Shoal Creek 393 454 on 389 442 26 208 244 Wills Crook yaa a 24 14 165 18 214 160 ‘Unit hydrograph obtained from the method of moments. ‘JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION ‘Normalized absolute sum of errora between observed and simulated unit hydrograph. [Normalized difference between peaks of observed and simulated unit hydrographs. ‘Unit hydrograph obtained from nonlinear constrained optimization, JAWRA Bovfadel closer to the parent peak flow than GRG2 though nei- ther method simulated very well the spiked peak on the unit hydrograph. Plots of the unit hydrographs for all five water- sheds appear in Figures 2a through 2e. At the Good- water watershed, there is a marked difference between the peak times of the parent hydrograph and the two derived hydrographs. This might reflect the inability of the Nash model to include channel trans- lation (Dooge, 1959). A UH with a narrow peak results from small storage capacity of the correspond ing watershed; this means the watershed behaves like ‘a channel. Hence, the UH (which is the direct runoff resulting from one rainfall pulse) will respond rapidly to rainfall by increasing sharply when rainfall occurs, and decreasing rapidly afterwards (j.e., when rainfall ceases). Overall, based on F,, Fz and plotted results, the Nash IUH appears capable of providing a repre- sentative hydrograph for the watersheds in question, especially if the parameters are estimated with non- linear constrained optimization, Phase 2: Unit Hydrograph Duration The situation considered during this phase more closely resembles what one would normally encounter when attempting to derive a UH of short duration from a parent UH of long duration. Results from Phase 2 are summarized in Table 2 where two inter- esting features stand out. First, in comparison to results for GRG2 in Phase 1, the error functions F, and F2 inerease. This implies that hydrographs esti- mated by nonlinear optimization do not match the shape or the peak of the parent hydrographs in Phase 2 as well as they did in Phase 1. This drop in perfor- ‘mance can be attributed to errors in the model or nat- ural noise in the data or both. For instance, the Nash IUH is unable to replicate the flashy hydrograph for Shoal Creek as shown in Figure 3d. This observation is consistent with Haktanir and Sezen (1990) who concluded that the gamma function cannot simulate narrow peaked unit hydrographs. Other sources of error in the conceptual model could result from quasi- linear rainfall runoff response or from noise in the runoff data. Second, the unit hydrographs found by linear inter- polation performed better than those given by the method of moments approach (Table 2) at all locations except the Potomac. In fact, at two of the watersheds (Shoal Creek and Wills Creek), the interpolated hydrographs fit the parent hydrograph better than the GRG2 hydrographs do. This is evident in Figures 3a through 3e, which show plots of the derived unit hydrographs and the parent hydrographs. In the examples presented here, the unnatural stepped JAWRA 172 nse of the interpolated unit hydrographs leads to ly minor discrepancies with the observed UH. Its interesting to note, however, that the interpolat- ed unit hydrographs consistently underestimate the peak flow of the parent unit hydrograph. Similar to the Nash IUH, the interpolated hydrographs have their greatest difficulty matching the flashy response of Shoal Creek. Phase 3: Runoff Simulation Unit hydrographs are often used to simulate rain- fall runoff from design storms. To compare the perfor- mance of the derived unit hydrographs, a hypothetical 12-pulse design storm was applied to the four unit hydrographs available at each of the five watersheds. ‘The amount of rainfall is not important because the F, and Fp functions were normalized with respect to total rainfall. The relatively long duration of the storm event is intended to provide each unit hydrograph with ample time to integrate the individ- ual rainfall pulses into a total watershed response. The rainfall-runoff routing was carried out with HEC-1. Simulation results are summarized in Table 3 and shown in Figures 4a through 4e. In Table 3, no single method emerges as superior with regard to matching the shape of the response generated from the parent hydrograph. At Bradford and Goodwater, GRG2 provides the best fit; on the Potomac, MoM works well; at Shoal Creek and Wills Creek, the interpolation approach is best. When aver- aged over the five watersheds, the linear interpolation scheme produces the lowest overall percent error (11.4 percent), followed by GRG2 (14.5 percent) and then ‘MoM (16.8 percent). All three derived hydrographs performed poorly on Shoal Creek in comparison to the parent hydrograph. ‘The magnitude of the F} error scores given in Table 3 for the simulated direct runoff hydrographs are con- sistently less than the F errors in Table 2 for the three estimation methods. The apparent improvement in performance can be attributed to the averaging which occurs when multiple rather than single inputs are convoluted with the unit hydrograph. This point is illustrated further in Figures 4a through 4e where multi-pulse storm hydrographs appear smoother than the derived single pulse unit hydrographs given in Figures 2 and 3. Hydrograph smoothing is expected since the rainfall-runoff model is a convolution opera- tion which integrates over the fluctuations in the unit response (Bruen and Doge, 1984), In Table 3 there is again no clear winner among the simulated hydrographs in matching the peak dis- charge obtained from the parent hydrograph. The GRG2 approach worked best at Bradford, Potomac JOURNAL OF THE AMERICAN WATER FIESOURCES ASSOCIATION Unit Hydrographs Derived from the Nash Model aot . ose — Parent oms ones 016 —— Moment aw x0 ox = 0025 I = 2 2 010 E 002 XN é = h Zo 5 oon) ff SY oon } cs vows ff x 008 ~ ous | [7 SX aoe tia oS 1 mm ms Time (6 increment) Time (05-hr increment) Figure 22, Bradford Watershed Phase 1 Figure 2b. Goodwater Watershed Phase 1 Parent and Derived Unit Hydrographs. arent and Derived Unit Hydrographe Pacont os — Parent oe E Graz rez i —— Moment = Moment 7 oo _ 8 2 Z £ é = oe Eo 3 i 3 J j | oor} oa J y ooo aot oo: 4 8 8 0 ow Oe o 2 # 6 8 0 8 Time (4-hr inerement) igure 2c. Potomac Watershed Phase 1 Parent and Derived Unit Hydrographs, 004 003 UH (thou) oot 02 4 8 8 wR Time (3-he increment) JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION Time (0.5 increment) Figure 2d. Shoal Creck Watershed Phase 1 Parent and Derived Unit Hydrographs, a Grcz | Figure 2e. Wills Creek Watershed Phase 1 Parent and Derived Unit Hydrographs, 173, JAWRA Boufadel 2.040 - — — Parent — Parent oss, GRe2 ou Interpolated 0.000 012 = 005 300 E om = ow = = 5 oars 5 00s e010 00 0.008 002 0.000 0.00 ~ o 2 # 6 8 re ee ee “Time (6-hr increment) Time (0.5-r increment) Figure Se. Bradford Wate: hed Phaso 2 igure 3b. Goodwater Watershed Phase 2 ‘Parent and Derived Unit Hedrographs. Parent and Derived Unit Hydrographs. — Parent Ee — Parent GRG2 GRG2 —— Intorpotatea| ry = zo 5 5 02 = 00 0 2 4 6 8 0 2 Mm 8 Oe o 2 4 Bll ‘Time (4 increment) Figure Se. Potomac Watershed Phase 2 Parent and Derived Unit Hydrographs, 005 003 UH (throu) 002 oot ‘Time (3-he increment) JAWRA ‘Time (0.5-hr increment) igure 34. Shoal Crock Watershed Phase 2 Parent and Derived Unit Hydrographs, Figure Se. Wills Creek Watershed Phase 2 Parent and Derived Unit Hydrographs. ‘JOURNAL OF THE AMERICAN WaTER RESOURCES ASSOCIATION ‘Unit Hydrographs Derived from the Nash Modet 2 . — Puet fe Gree = 7° = ted at 8 cs 8 ge ee z 3 : e es 5 : : 7 - see ee Tino (4 erent) Tino (tv eromat Figaro. Bred tered Phan § Figure 4 Gendeater Watershed Phase 3 Pien tabettod Ua Roser eta ta Betel Uaibowes 5 55 . 4s — Parent — Parent Flow (1000 fs) ° 5 0 6 Ed 2s Time (4-he increment) Figure 4c, Potomac Watershed Phase 3 Parent and Derived Unit Hydrographs. 160 0 120 Flow (1000 cfs) 5S SO Time (3-hr increment) JOURWAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 175 Flow (1000 ets) ‘Time (3-r increment) igure 4d. Shoal Creek Watershed Phase 3 Parent and Derived Unit Hydrographs, Figure 4e, Wills Creck Watershed Phase 3 Parent and Derived Unit Hydrographe. JAWRA Errore in the Simulated Direct ‘Runoff Shape: F; (percent) ig (percent) Hydrograph GRE? MoM Interpolated GRG2 MoM Interpolated Bradford 751 122 92 180 “138 4.05, Goodwater Ms 208 Mar 462 “164 212 Potomac 108 8.55 931 3.10 7.10 165 ‘Shoal Creek m2 312 m2 120 299 16a Wille Crock 124 no 687 133 107 108 Normalized absolute num of errors between observed and simulated direct runoff Normalized difference between peaks of observed and simulated direct runoff. Unit hydrograph obtained from nonlinear constrained optimization Unit hydrograph obtained from the method of moments. Unit hydrograph obtained by linear interpolation. and Shoal Creek; MoM performed well at Wills Creek; the interpolated hydrograph provided the best match to peak flow at Goodwater. When averaged over all five watersheds, GRG2 achieved the lowest percent peak flow error (6.9 percent), followed by linear inter- polation (8.2 percent) and then by MoM (15.6 per- cent). Shoal Creek continued to be the location with the largest relative error. Here too, as in Table 2, the interpolation approach consistently underestimated the magnitude of the peak flow. Comparing Tables 2 and 3 shows that the method of moments improved its performance at all five sites; however, no pattern is discernible with the other two derived unit hydro- graphs. CONCLUSIONS Many hydrologists have found that the gamma dis tribution provides a reasonable model for the unit hydrograph (Edson, 1951; Williams, 1968; Aron and White, 1982; Bruce and Brown, 1992). This paper fur- ther examined the suitability of the gamma model for simulating watershed response. In particul focused on the Nash IUH as a tool for de short duration unit hydrographs from a parent unit hydrograph of long duration. It has contrasted two methods for estimating the parameters of the Nash TUH (n and k) and compared the performance of the Nash model against an empirical interpolation scheme for deriving short duration unit hydrographs. The key findings include: 1. The Nash IUH is a reasonable model for simu- lating watershed response on a continuous basis when the parent unit hydrograph has a mild shape. JAWRA In this case, the Nash IUH provides a sound and practical approach for generating short duration unit hydrographs from long duration parent unit hydro- graphs. 2. The Nash IUH, being a gamma function, cannot simulate narrow fast-peaking unit hydrographs due to its inability to include channel translation. 3. Changing hydrograph duration does not affect considerably estimates of the Nash IUH parameter pair (n,k) obtained from nonlinear constrained opti- mization. 4, Short duration unit hydrographs obtained from the Nash IUH performed best for reproducing the peak flow of the parent hydrograph in comparison to linear interpolation which consistently underestimat- ed the peak flow. 5. When simulating runoff events resulting from multi-pulse inputs, differences in the total watershed response produced by various derived unit hydro- graphs tend to diminish due to averaging effects of the rainfall-runoff convolution operation. One of the problems in determining the best fit between a derived unit hydrograph and an observed parent unit hydrograph is the difficulty in quantifying hydrograph errors in one number. In this study we defined two types of hydrograph disparity. One mea- sured the agreement between hydrograph shapes (F,) and the other measured agreement between peak flows (F,). Many other disparity criteria could be developed. In practice, the choice of which error crite- rion to adopt depends on need. If the objective of the hydrologist (or watershed manager) is to estimate watershed abstractions or some other continuous pro- cess, then the shape criterion would be the most appropriate way to measure goodness of fit between hydrographs. Conversely, if his/her objective is to 176 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION Unit Hydrographa Derived from the Nash Model ‘estimate the peak flow, then the peak criterion would bbe the relevant measure. ‘Overall, the Nash IUH model offers a versatile and convenient tool for modeling rainfall runoff from watersheds where the response is not too flashy. The practical convenience and sound theoretical basis of the linear convolution approach assure continued application of the Nash IUH in watershed modeling. ACKNOWLEDGMENTS: suthor is grateful to his Masters thesis advisor, Dr. Steven Buchberger of the University of Cincinnati for guidance, The ‘author is also grateful to the Miami Conservancy District for finan- al support and for providing date from the Bradford, Ohio, water- ‘shed and to Dr. Allen Hjolmfelt of the USDA-ARS for providing data from the Goodwater, Missouri ste. LITERATURE CITED ‘Abramawitz, M. and I. A. Stogun, 1972. Handbook of Mathematical ‘Functions. National Bureau of Standards, U.S. Department of Commerce, Washington, D.C. ‘Aron, G and B. L. White, 1982, Fitting a Gamma Distribution Over 'r Synthetic Unit Hydrograph. Water Resources Bulletin 181) 95.98. Beck J. V. and K, J. Amold, 1977. Parameter Estimation in Engi- ‘neering and Science, John Wiley, New York, New York, pg. 501. Blank, D._ J. W. Delleur, and A. Giorgini, 1971. Oscillatory Kernel Functions in Linear Hydrologic Models. Water Resources Research 1(5):1102-1117. Boufadel M.C., 1992, Discrete Nash Model to Simulate Continuous ‘Unit Hydrographs,, Masters Thesis, University of Cincinnati Cincinnati, Ohio. Bruce, N. W. and J. W. Brown, 1902. Development and Evaluation of a Dimensionless Unit Hydrograph. Water Resources Bulletin 26(2):397-408. Bruen, M. and J. C. Dooge, 1984, An Efficient and Robust Method Yor Estimating Unit Hydrograph Ordinates. Journal of Hydrolo- sy 10:1-24. Chow, V. T., D. R. Maidment, and L, W. Mays, 1988. Applied Hydrology. McGraw Hill, New York, New York, 572 pp. Diskin, M. HL, 1969, Evaliation of Segmented [UH from Deriva ‘tives, Proceedings ASCE Hydraulics Division (HY1):320-046. Dooge, C. I. J. 1969, A General Theory of the Unit Hydrograph. ‘Journal Geo Research 64(2):241-256. Eagleson, P. ‘and F. March, 1966. Computation of ‘Optimum Realiz Rescarch 2(4):765-764, Eason, C. G., 1961, Parameters for Relating Unit Hydrographs to ‘Watershed Characteristics, Transactions of the American Geo- physical Union 82:591-594. Gray, D. M., 1961, Synthetic Unit Hydrographs for Small Water- sheds. Proceedings ASCE Hydraulics Division (HY): 33-54 Haan, C. T, 1977, Statistical Methods in Hydrology. Iowa State ‘University Press, Ames, Iowa, 378 pp. Haktenir, T. and N. Sezen, 1990. Suitability of Two-Perameter ‘Gamma and Throo-Parameter Beta Distributions as Synthetic ‘Unit Hydrographs, Hydrological Sciences 35(2):167-184, ‘JOURNAL OF THE AMERICAN WATER Rlesounces AssocuTion 177 Hielmfelt, A.T. and L. A. Kramer, 1988. Unit Hydrograph Varisbili- ty for « Small Agricultural Watershed. Proceedings ASAE Sym- posium on Modeling Agricultural, Forest and Rangeland Hydrology, Chicago, Ilinois, pp. 57-366. Lasdon, L. 8 A. D. Waren, A. Jain, and M. Ratner, 1978. ACM "Transection Mathematical Software 4(1):34-60. Mays, L. W. and 8. L. Coles, 1980. Optimization of Unit Hydro- graph Determination, ASCE Journal of Hydraulic Division, 106(FY):85.97. ‘Mays, L. W. and C. K Taur, 1982, Unit Hydrographs Via Nonlinear ‘programming, Water Resources Research 18(4):744-752. Nash, J. E., 1967, The Form of the Instantaneous Unit Hydro. ‘graph. International Association of Science and Hydrology S:1l4i21. Nash, JE., 1958. Systematic Determination of Unit Hydrograph ‘Parameters. Journal of Geophysical Research 64(1):111-11. Newton, D. W. and J. W, Vinyard, 1967. Computer Determined Unit ‘Hydrographs from Floods, ASCE Journal of Hydraulics Division 99(5):219-234. Sangal, B. P, 1986, Recursion Formula for Unit Hydrographs ‘Canadian Journal of Civil Engineering,13:386-388. Sherman, L. K., 1932, Streamflow from Rainfall by Unit Graph “Method. Engineering News Record, pp. 501-508 Singh, K.P, 1976, Unit Hydrographs-a Comparative Study. Water ‘Resources Bulletin 12(2):381, Singh, V. P, 1992. Elementary Hydrology. Prentice Hall, New York, New York. ‘Tauxe, G. W, 1978. S-Hydrographs and Change of Unit Hydro- raph Duration. ASCE Journal of Hydraulics Division 1OACHYS):439-444, Unver, O and L, W. Mays, 1984. Optimal Determination of Loss ‘Rave Functions and Unit Hydrographs. Water Resource Research 20(2):208-214, US. Army Corps of Engineers, 1990. HECI-Users Manual. Davis, California Williams, J. 1968. Runoff Hydrographs from Smell Texas Black- lands Watersheds, US. Department of Agriculture, Agriculture Research Service, ARS 41-143, JAWRA

You might also like